摘要
巡检机器人提供了海量的设备实时运行图像及声音数据,而常规数据处理方式通常受限于固定的数学处理工具,难以准确去除图像噪声、识别出设备异常声音,因此不能获得有效、可靠的设备运行数据。针对此问题,采用深度学习对运行设备的实时图像及声音信息进行数据分类等预处理并获得典型训练集。在此基础上,进行大量训练并获得训练模型,将训练出的模型应用于实时图像及声音,从而提取出有效的设备运行数据。Matlab仿真结果表明,深度学习能更优地提取数据特征,较好地去除图像噪声并精确识别设备的异常声音,解决了传统机器学习缺乏训练数据、泛化能力不足的问题。
The inspection robot provides a large amount of real⁃time running image and sound data of equipment,while the conventional data processing methods are often limited by the fixed mathematical processing tools,which are difficult to accurately remove image noise and identify the abnormal sound of equipment,so it can not obtain effective and reliable data of equipment operation.In order to solve this problem,this paper decided to use deep learning to preprocess the real⁃time image and sound information of the running equipment and obtain the typical training set.On this basis,a lot of training was carried out to obtain the training model.The trained model was applied to the real⁃time image and sound to extract the effective operation data of the equipment.Matlab simulation results show that deep learning can better extract data features,remove image noise and accurately identify abnormal sound of equipment,and solve the problem of lack of training data and generalization ability of traditional machine learning.
作者
李标俊
姚传涛
杨贵军
桂辉阳
毛臻炫
凌艺榕
LI Biaojun;YAO Chuantao;YANG Guijun;GUI Huiyang;MAO Zhenxuan;LING Yirong(Tianshengqiao Bureau of UHV Transmission Company,China Southern Power Grid Corporation,Xingyi 562400,China)
出处
《电子设计工程》
2020年第23期68-72,共5页
Electronic Design Engineering
基金
中国南方电网有限公司深圳供电局有限公司科技项目(090000GS62161590)。
